Electricity + Control February 2017

CONTROL SYSTEMS + AUTOMATION

Weather Forecasting Meets Sophisticated Analytics

Robbie Berglund, The Weather Company

Energy and utilities sectors are weather dependent industries and weather can affect domestic load, commercial load and public load, not to mention operations, efficiency and safety.

W eather can potentially impact every person, and every business, on the planet, every day. When a company’s profitability is dependent on weather, accuracy and insight can be paramount to success… not to mention the effect weather can have on utilities and industry. An inexact − but critically important − science Historically, load forecasting − in essence, predicting utility demand and consumption − has been a complex and uncertain process. The ability to accurately forecast load can help inform mission-critical decisions across all operations, from electric power generation and purchasing, to load switching, infrastructure and even staffing. In fact, forecasting, whether it’s effective or not, can have ramifications for all entities involved in energy generation, transmission, distribution, marketing and financing. One reason load forecasting has been challenging is that there are multiple variables to take into account. These include time (hour of the day, day of the week, weekday vs. weekend, and holidays); popula- tion usage (types of customers, increased or decreased numbers of customers, and changes in usage); special events (local, national or international); and current, recent or projected energy prices. That said, weather is arguably one of the most important pieces of the puzzle. Sunny with a chance of increased load Extreme weather is often referred to as ‘an act of God’. No one can predict the weather with absolute certainty. But, weather conditions can significantly influence load, which in turn, may significantly influ- ence performance and profitability. Variables such as temperature and humidity have a direct correlation with energy consumption for cooling and heating. Two standard industry measures, THI (Temperature-Humidity Index) and WCI (Wind Chill Index) are used by most utility compa- nies. But other variables are important as well. Visibility, precipita- tion and cloud cover can also affect consumption. As can whether

temperatures are above- or below-average, and how long a particular heat wave or cold snap lasts. Quite simply, we believe accurate load forecasting depends on accurate weather forecasting.

Leveraging accurate weather forecasts and data analytics

At The Weather Company (further referred to as ‘the company’), an IBM Business, significant investments have recently been made in both: • An improved weather forecasting system • Data science capabilities The resulting system was designed to create an industry leading product that provides accurate, timely, and spatially resolute weather forecasts while expertise in the latter allows us to convert these ac- curate weather forecasts into user-friendly products for clients in the utility and energy trading businesses. The Load Forecast feature of our flagship, WSI Trader, is anchored in advanced and proprietary weather and data science. In our experi- ence, good load forecasts are strongly dependent upon good weather forecasts. The company’s weather forecasting engine (Forecasts on Demand, or FoD) is an automated system that produces hourly fore- casts for all of the most relevant weather variables (e.g., temperature, dew point, wind speed, precipitation, cloud cover, snowfall) at 4- m spatial resolution across the globe, allowing for hyper-local insight – of particular value to ISOs. Improved models can help improve load forecasting Thes company’s FoD forecasts are a skill-weighted blend of available weather models, including the ECMWF, GFS, and NAM models (de- terministic and ensemble), along with GFS MOS and the company’s proprietary high-resolution weather model (RPM). Weights are assigned to each model based on the optimal combi- nation of bias-corrected model forecasts over the most recent weeks. The first few hours of the forecast period are ‘forward-corrected’ based upon the latest observations.

Electricity+Control February ‘17

16